The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest data from a wide variety of external sources, including business intelligence dashboards, code repositories, team communication channels, and documentation platforms. This data is processed to generate and update semantic layer artifacts.
Ingestion points: Data is gathered via connectors and tools from sources like Slack, GitHub, and various data warehouses (as described in references/semantic-layer/connector-playbook.md).
Boundary markers: The instructions in references/semantic-layer/weekly-polling-automation.md establish boundaries for updates, differentiating between canonical sources and observed behavior.
Capability inventory: The skill possesses the capability to write files to specific local directories and execute source-reading tools through the agent's platform.
Sanitization: Explicit instructions are provided in SKILL.md and references/semantic-layer/connector-playbook.md to exclude credentials, sensitive personal data, and row-level examples from generated artifacts.
Local Filesystem Persistence: A core function of the skill is to write and maintain semantic layer files within the $CODEX_HOME or ~/.codex directories. This allows the data context to be persisted and referenced in future sessions. These file operations are restricted to the designated skill and state directories.
Subprocess Execution in Testing: The unit tests included with the skill use the Python subprocess module to run internal helper scripts (data_analytics_preflight.py and record_plugin_install_suppression.py). This execution is limited to the skill's own scripts within the test environment for validation purposes.

create-data-context